Divergence Between StageIII and StageIII/VIa Control Arms Explained
Frugal N: The “divergence” you notice is correct.
Recall that we know one thing for sure: 298 enrollees died. We don’t know which arm. So the possible StageIII/IVa control arm has lower overall survival (higher morbidity) than a StageIII-only control arm that has higher survival (lower morbidity). We don't know what cancer type was finally enrolled, likely a mix of the two. So these two survival curves are our attempt to establish “guardrails,” to speculate about the nature of the actual survival function for the test arm.
So... if you subtract 298 from either control arm possibility, the remaining deaths had to occur in the corresponding test arm. If we observe in the P3 study that the control indeed was comprised of the more deadly StageIII/IVa cancers, then the study’s test arm MUST have less morbidity (higher survival) than if the control arm consisted of only the less-deadly StageIII.
My purpose in designing our calculator with two different control arms was to compare the range of overall survival based upon differing assumptions for the control group survival.
Much has been discussed about the clinical trial effect. Not so much about the enrollee severity effect. Perhaps those with more advanced cancers and their doctors were more willing to “try anything.” Could the study be packed with more severe cases? If so, then the OS we observe could significantly surpass expectations.
I urge you to mess around with the inputs. It doesn’t take long to get a “feel” for inferred OS under various scenarios. Hope this helps.
Cheers from a cloudy Oregon,
Jim
